Artificial intelligence–based prognostic modeling of immunoradiotherapy in Barcelona clinic liver cancer stage C hepatocellular carcinoma: a multicenter retrospective study

Background: Barcelona Clinic Liver Cancer (BCLC) stage C hepatocellular carcinoma is associated with poor prognosis, and conventional systemic therapies offer limited survival benefit. Immunotherapy combined with radiotherapy has emerged as a promising approach, but patient responses are heterogeneous. Artificial intelligence (AI) may facilitate individualized prognostic prediction to guide therapy. Methods: We retrospectively analyzed 198 BCLC stage C HCC patients from three centers. The experimental group received immunoradiotherapy plus targeted therapy, and the control group received immunotherapy plus targeted therapy. Baseline characteristics were balanced using inverse probability of treatment weighting (IPTW). Five machine learning models (Cox, LASSO, DT, RSF, and XGBoost) were developed to predict 6-, 12-, and 24-month overall survival. Results: Before and after IPTW adjustment, the experimental group showed longer progression-free and overall survival than the control group. In the training cohort, the RSF model achieved the highest concordance index (0.7458). In the validation cohort, it also demonstrated the best receiver operating characteristic - area under the curve (ROC-AUC) values for 6-, 12-, and 24-month OS (0.821, 0.818, and 0.791, respectively). Decision curve analysis and calibration plots indicated good stability. Variable importance analysis showed that tumor number, tumor size, and portal vein tumor thrombosis consistently contributed substantially to survival prediction across all time points. Conclusions: Immunoradiotherapy represents a promising therapeutic option for BCLC stage C HCC. The RSF-based model may support individualized prognostic risk stratification and clinical decision-making.